Allen Riddell

I find myself doing similar operations. @JanSchulz have you tried to express some of your functions in terms of functions available in `toolz` (`toolz.dicttoolz`) or `funcy`? There's a lot of...

These are good ideas! I still think HCA or Mallet is probably the right tool for serious fitting but I'm happy to implement these changes. For simplicity's sake and to...

re: random_state the issue is you want to record the seed passed somewhere so someone can look at it (in case they forgot it) and then rerun the model to...

You make some good points. Can I think about this more in a week or so when I have more time? In principle, I'd really like to stick to the...

Maybe one could do something innovative here. If one ran two threads in parallel one could test for convergence, I think. `n_iter` could become `min_iter` and `max_iter` but one could...

Processing requirements will never be a problem. Everyone has more than 1 core these days; one can release the thread in Cython. WS and DS can be shared by both...

Test for convergence by calculating the Rhat of the complete loglikelihood of the two chains. Perhaps one can do this ever 100 iterations.

I've had another thought. What if one passed data from the existing model (e.g., the phi and the theta) as initial values to a fresh fit. It would be continuation...

I mean the second. One would add new parameters like initial_theta and initial_phi (or initial_ndz initial_nzw) to **init**. Multiple cores is a great idea. Someday I'll get around to this....

This isn't trivial but it seems possible. The strategy looks like to use a hook to generate the appropriate `.c` files using cython and then build using pbr as if...